Introduction

The fishHook R package enables agile statistical exploration of coding and non-coding mutational recurrence in cancer through generalized linear modeling (GLM) of heterogenous neutral somatic mutation rates along the genome. fishHook can be applied to the analysis of genes, enhancers, promoters, genomic tiles, or any arbitrary “hypothesis set” (defined by the user as a set of genomic intervals or GRanges), as well as complex sets of the above (e.g. genes sets representing pathways, enhancers sets known to interact with a target gene). The core of the fishHook approach is to enable nomination of loci following the correction of known covariates of neutral mutation, including chromatin state, replication timing, and nucleotide context. The goal of fishHook is to identify cancer “drivers”, i.e. loci that are under positive somatic selection through the application of an accurate null / background model whose application yields straight Q-Q plots.

Though we provide many pre-computed covariates, the key power of fishHook is to empower the user to add custom covariates and (more broadly) provide a framework for generation of custom models, e.g. fitting variant- and tumor-type specific linear models and then integrating deviations from these models to nominate loci (or sets of loci) that deviate from the NULL.

Please include this citation if you use fishHook: Imielinski, Guo, Meyerson Cell. 2017 Jan 26;168(3):460-472

Driver discovery in cancer whole exomes

We will demonstrate a quick whole exome analysis using public TCGA lung adenocarcinoma mutation data. Additional packages like gTrack and rtracklayer will help with data import and visualization, but are not necessary to run fishHook.

library(fishHook)    
library(gTrack)
library(plotly)
library(rtracklayer)

Read in data

Read in the mutation data and additional tracks that we will use in this analysis.

## mutation calls cached from public GDAC Broad firehose https://bit.ly/2sFxWY6
mutations = dt2gr(fread('http://mskilab.com/fishHook/hg19/luad.maf'))

## GENCODE v19 genes these are our "hypotheses"
genes = gr.sub(import('http://mskilab.com/fishHook/hg19/gencode.v19.genes.gtf')) %Q% (gene_type == 'protein_coding') %Q% (level<3)

## protein coding cds definitions
cds = readRDS(gzcon(file('http://mskilab.com/fishHook/hg19/gencode.v19.cds.rds')))

## bigWig file of fractional coverage of hg19 positions by Agilent exome
## we will use this in combination with cds to define eligible positions
exomecov = import('http://mskilab.com/fishHook/hg19/exome_coverage.bw')

Take a peek at our mutations

head(mutations[, c('Tumor_Sample_Barcode', 'Variant_Type', 'Variant_Classification', 'Reference_Allele', 'Tumor_Seq_Allele2')])
## GRanges object with 6 ranges and 5 metadata columns:
##       seqnames    ranges strand |         Tumor_Sample_Barcode
##          <Rle> <IRanges>  <Rle> |                  <character>
##   [1]        1    905907      * | TCGA-05-4249-01A-01D-1105-08
##   [2]        1   1192480      * | TCGA-05-4249-01A-01D-1105-08
##   [3]        1   1854885      * | TCGA-05-4249-01A-01D-1105-08
##   [4]        1   9713992      * | TCGA-05-4249-01A-01D-1105-08
##   [5]        1  12908093      * | TCGA-05-4249-01A-01D-1105-08
##   [6]        1  17257855      * | TCGA-05-4249-01A-01D-1105-08
##       Variant_Type Variant_Classification Reference_Allele
##        <character>            <character>      <character>
##   [1]          SNP      Missense_Mutation                A
##   [2]          SNP                 Silent                C
##   [3]          SNP                    IGR                G
##   [4]          SNP                 Intron                G
##   [5]          SNP      Missense_Mutation                C
##   [6]          SNP      Missense_Mutation                C
##       Tumor_Seq_Allele2
##             <character>
##   [1]                 T
##   [2]                 A
##   [3]                 C
##   [4]                 A
##   [5]                 A
##   [6]                 T
##   -------
##   seqinfo: 25 sequences from an unspecified genome

Instantiate FishHook object from events and eligible territory

First we define the “eligible territory”. This is a key component of all somatic mutational recurrence analyses, since not all of the genome is covered by most sequencing assays. For example in a whole exome sequencing dataset, less than 2% of the genome is reliably captured. In a targeted sequencing panel, this fraction will be even smaller. Even in whole genome sequencing using Illumina short reads, only a subset (70%) of the genome is reliably callable (Heng Li Bioinformatics 2014 Oct 15;30(20):2843–51)

Eligible territory coverage will influence the “denominator” of our recurrence analysis, i.e. the number of positions in each hypothesis interval where a mutation could have possibly been detected. If we do not take eligible territory into account we will mis-estimate the background mutation rate in a given region.

To define eligible territory for this whole exome analysis, we will choose the portion of cds (protein coding) bases that are captured in at least 95% of whole exomes, which represents about 26MB of genome.

eligible = reduce(intersect(exomecov %Q% which(score>0.95), cds, ignore.strand = TRUE))

We define “events” as nonsynonymous mutations. In this simple model, we will lump together SNVs (of different flavors), and indels (of different flavors). (We discuss more complex models that subdivide mutation types later in the tutorial).

events = mutations %Q% (Variant_Classification != 'Silent')

Now that we have loaded our hypotheses (genes), events, and eligible, we are ready to create and analyze a basic FishHook object. Instantiation of the object will involve counting how many events are in the eligible portion of each hypothesis interval. We provide the “idcol” parameter so that each tumor sample (as defined by the “Tumor_Sample_Barcode” column in this events GRanges) can only provide at most one event to the counts in each interval.

fish = Fish(hypotheses = genes[, 'gene_name'], events = events, eligible = eligible, idcol = 'Tumor_Sample_Barcode')

## Contains 19796 hypotheses.
## Contains 56809 events to map to hypotheses.
## Spanning 24.58 MB of eligible territory.
## Covariates:
## 
## Hypotheses contains 1 metadata columns.
## Current State: Annotated

Run basic FishHook model without covariates

We can score this basic FishHook object, i.e. compute p values for every hypothesis, using a simple glm that models a uniform mutation density along the genome, i.e. the glm fits only an intercept and applies no covariates (after correcting for the number of eligible bases in each interval).

The $res field of the FishHook contains a data.table of scoring results. $res has one row per input hypothesis, with p values, FDRs, and additional interval annotations provided with the hypotheses GRanges.

fish$score()
head(fish$res[order(p), ])
fish$qqp()

seqnames	start	end	width	gene_name	p	fdr	count	effectsize	count.pred	count.density	count.pred.density	p.neg	fdr.neg
17	7565097	7590856	25760	TP53	0.0e+00	0.0000	106	5.62	2.15	0.0941	0.00191	1	1
2	79384132	79386879	2748	REG3A	1.0e-07	0.0009	17	4.09	1.00	0.0324	0.00191	1	1
19	10596796	10614417	17622	KEAP1	5.0e-07	0.0031	30	3.54	2.57	0.0222	0.00191	1	1
5	24487209	24645087	157879	CDH10	3.6e-06	0.0180	41	3.19	4.50	0.0174	0.00191	1	1
3	147103833	147124647	20815	ZIC4	5.5e-06	0.0200	18	3.41	1.69	0.0203	0.00191	1	1
8	88882973	88886296	3324	DCAF4L2	6.4e-06	0.0200	22	3.28	2.26	0.0186	0.00191	1	1

You will notice that this Q-Q plot appears curved and inflated, though its slope (lambda) is reasonably near 1. The low alpha value (MLE of the alpha parameter of the Gamma distribution), suggests that the GLM is detecting additional variance in the data that is unmodeled by a pure Poisson regression. Adding covariates to the model should improve the quality of the fit.

The top hits in the plot (you can hover over them) include TP53 but also many unlikely cancer gene candidates. Among these are olfactory receptors, which are located in late replicating regions of the human genome and thus accumulate neutral mutations more frequently (Lawrence et al 2013 Nature Jul 11;499(7457):214-218).

Add covariates to FishHook model

To address these issues, we will load in data specifying replication timing, chromatin state, and nucleotide context. These are all important determinants of somatic neutral mutation density. We load these data in as GRanges objects using functions in data.table, rtracklayer, and gUtils packages (however you are free to use any GRanges import utility of your choice).

We first load in replication timing data as a GRanges then instantiate it as a Covariate Replication timing information is contained in the $score metadata field of reptime. We instantiate it as a covariate of type “numeric” by specifying field score.

## replication timing for NHEK obtained from  https://bit.ly/2sRsXT9 and
## converted to rds via rtracklayer::import
reptimedata = readRDS(gzcon(file('http://mskilab.com/fishHook/hg19/RT_NHEK_Keratinocytes_Int92817591_hg19.rds')))

reptime = Cov(data = reptimedata, field = 'score')

Below, context is a GRanges object with 98 columns representing tri, di, and mononucleotide context counts in the hg19 genome. Code for computing context (e.g. for another genome) is provided here.

We instantiate a numeric Covariate object from context, choosing only two of the columns here to take into account G and C content. Note that the covariate object can be vectorized (concatenated, subsetted) and instantiated around several columns of an input GRanges. As a result, contextcov will be length 2 (representing G and C nucleotides fraction).

context = readRDS(gzcon(file('http://mskilab.com/fishHook/hg19/nucleotide.context.rds')))
gc = Cov(data = context, field = c('C', 'G'))

Finally we load in chromHMM data for cell line A549 from Epigenomics Roadmap. We will want to create a covariate that will model the fraction of heterochromatic and quiescent regions in each query interval.

To do so, we will create an “interval” covariate by not specifying a metadata field.

### data cached from https://egg2.wustl.edu/roadmap/data/byFileType/chromhmmSegmentations/ChmmModels/coreMarks/jointModel/final/E114_15_coreMarks_mnemonics.bed.gz
hetchromdata = gr.sub(import('http://mskilab.com/fishHook/hg19/E114_15_coreMarks_mnemonics.bed.gz'), 'chr', '') %Q% (name %in% c('8_ZNF/Rpts', '9_Het', '15_Quies'))
hetchrom = Cov(hetchromdata, name = 'Heterochromatin')

We now add these covariates to the model. For type numeric covariates, e.g. replication timing, this will trigger the calculation of the average value of each covariate within the eligible subset of each hypothesis interval. and the fractional overlap of of its eligible subset This annotation is the most computationally intensive and slowest aspect of fishHook analyses, though occurs within a few seconds for this small number of covariates.

fish$covariates = c(reptime, gc, hetchrom)
## FishHook 2018-06-11 19:49:06: Aggregating covariates over eligible subset of hypotheses
## FishHook 2018-06-11 19:49:06: Overlapping with covered intervals
## FishHook 2018-06-11 19:49:06: Finished overlapping with covered intervals
## FishHook 2018-06-11 19:49:06: Annotating track score
## FishHook 2018-06-11 19:49:11: Annotating track C
## FishHook 2018-06-11 19:49:13: Annotating track G
## FishHook 2018-06-11 19:49:16: Annotating track Heterochromatin
## Warning in self$clear(): Resetting scores since covariates re-defined,
## please run $score() to get updated p values

Now that we’ve added covariates, we can rescore fish and compute p values. Looking at these results, we immediately see an improvement in lambda (closer to one) and alpha (increased), and the nominated gene list (no more olfactory receptors, We see also a reasonable number of significant (fdr<0.1) genes at the top of the list (or at the top right of the QQ plot) that have been biologically implicated in lung adenocarcinoma tumorigenesis.

fish$score()
fish$res[order(p)[fdr<0.1, ]
fish$qqp()

seqnames	start	end	width	gene_name	p	fdr	count	effectsize	count.pred	count.density	count.pred.density	p.neg	fdr.neg
17	7565097	7590856	25760	TP53	0.0e+00	0.00000	106	6.27	1.370	0.09410	0.00121	1	1
19	10596796	10614417	17622	KEAP1	0.0e+00	0.00000	30	4.02	1.840	0.02220	0.00137	1	1
7	55086714	55324313	237600	EGFR	1.0e-07	0.00054	44	3.06	5.280	0.01200	0.00144	1	1
7	142457319	142460923	3605	PRSS1	6.9e-06	0.03400	16	3.06	1.910	0.02160	0.00258	1	1
1	157800704	157868046	67343	CD5L	8.7e-06	0.03400	14	3.07	1.670	0.01340	0.00161	1	1
21	44513066	44527697	14632	U2AF1	1.8e-05	0.05400	8	3.38	0.768	0.01180	0.00113	1	1
14	19553365	19590078	36714	POTEG	2.0e-05	0.05400	13	2.99	1.640	0.02210	0.00279	1	1
11	62443970	62446567	2598	UBXN1	2.2e-05	0.05400	10	3.28	1.030	0.01140	0.00118	1	1
1	175284330	175712906	428577	TNR	2.8e-05	0.05700	38	2.49	6.740	0.00933	0.00166	1	1
6	136578001	136610989	32989	BCLAF1	2.9e-05	0.05700	23	2.66	3.650	0.00834	0.00132	1	1
7	140419127	140624564	205438	BRAF	3.6e-05	0.06500	21	2.66	3.320	0.00984	0.00156	1	1
8	10463859	10569697	105839	RP1L1	4.6e-05	0.07600	48	2.34	9.450	0.00791	0.00156	1	1
19	10416103	10426691	10589	FDX1L	5.3e-05	0.08100	4	4.02	0.247	0.01650	0.00102	1	1

Note that this model is still quite rudimentary (e.g. we have lumped together SNVs and indels, we have not substratified SNVs by mutational context, we have employed very few covariates) but we still get a reasonable gene list and minimal statistical inflation.

We can inspect the parameters of this model see which features it’s using. We can see from the estimate and Pr(|>z|) that replication timing is significantly negatively correlated and heterochromatin is significantly positively correlated with mutational density (as expected).

summary(fish$model)
## 
## Call:
## glm.nb(formula = formula, data = as.data.frame(tdt), maxit = iter, 
##     init.theta = 3.514844957, link = log)
## 
## Deviance Residuals: 
##     Min       1Q   Median       3Q      Max  
## -3.1965  -0.9862  -0.2834   0.4381  15.5196  
## 
## Coefficients:
##                 Estimate Std. Error  z value Pr(>|z|)    
## (Intercept)     -6.63330    0.04891 -135.631  < 2e-16 ***
## score           -0.51553    0.01023  -50.411  < 2e-16 ***
## C                0.70117    0.41074    1.707 0.087801 .  
## G                1.47342    0.41306    3.567 0.000361 ***
## Heterochromatin  0.33214    0.01672   19.860  < 2e-16 ***
## ---
## Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
## 
## (Dispersion parameter for Negative Binomial(3.5148) family taken to be 1)
## 
##     Null deviance: 23853  on 18186  degrees of freedom
## Residual deviance: 19088  on 18182  degrees of freedom
## AIC: 65310
## 
## Number of Fisher Scoring iterations: 1
## 
## 
##               Theta:  3.5148 
##           Std. Err.:  0.0935 
## 
##  2 x log-likelihood:  -65298.0760

Analyze truncating mutations and pathways

We can run a similar analysis but choosing only truncating mutations (by subsetting the mutation GRanges). Scoring this new model, we obtain a different mutation list, containing likely candidate drivers with enrichment of frameshift, nonsense, or nonstop indels and SNVs. This includes genes like ARID1A, RBM10, and SETD2.

## replace events with new subset of mutations
fish$events = mutations %Q% 
                 (grepl('(Frame_Shift_)|(Nonsense)|(OutOfFrame)|(Nonstop)', Variant_Classification))

## re-score model and inspect results
fish$score() 
head(fish$res[order(p), ])
fish$qqp()

seqnames	start	end	width	gene_name	p	fdr	count	effectsize	count.pred	count.density	count.pred.density	p.neg	fdr.neg
17	7565097	7590856	25760	TP53	0.0e+00	0.00000	33	7.72	0.1570	0.02930	0.000139	1	1
3	47057919	47205457	147539	SETD2	0.0e+00	0.00014	18	4.23	0.9620	0.00295	0.000157	1	1
X	47004268	47046212	41945	RBM10	1.0e-07	0.00038	6	5.52	0.1310	0.01140	0.000249	1	1
16	3115298	3131908	16611	IL32	1.2e-06	0.00590	4	5.61	0.0820	0.00905	0.000186	1	1
8	126442563	126450647	8085	TRIB1	3.7e-06	0.01500	4	5.48	0.0898	0.00531	0.000119	1	1
1	27022524	27108595	86072	ARID1A	4.8e-06	0.01600	10	3.84	0.6980	0.00176	0.000123	1	1

Read in and parse reactome pathways as list of gene symbols.

## parse Reactome pathways from .gmt format
pathways = strsplit(readLines('http://mskilab.com/fishHook/hg19/ReactomePathways.gmt'), '\t')
pathways = structure(lapply(pathways, '[', -1), names = sapply(pathways, '[', 1))

## match them to create sets of indices as a named list
sets = sapply(sapply(pathways, match, fish$hypotheses$gene_name), setdiff, NA)

Here is what the pathways and sets look like:

head(pathways[1:2])
## $`2-LTR circle formation`
##  [1] "R-HSA-164843" "BANF1"        "HMGA1"        "LIG4"        
##  [5] "PSIP1"        "XRCC4"        "XRCC5"        "XRCC6"       
##  [9] "gag"          "gag-pol"      "rev"          "vif"         
## [13] "vpr"          "vpu"         
## 
## $`5-Phosphoribose 1-diphosphate biosynthesis`
## [1] "R-HSA-73843" "PRPS1"       "PRPS1L1"     "PRPS2"
head(sets[1:2])
## $`2-LTR circle formation`
## [1] 10815  6342 12670  8606  5385  3035 18801
## 
## $`5-Phosphoribose 1-diphosphate biosynthesis`
## [1] 19432  7085 18957

The set analysis is a bit more computationally intensive. We can speed things up through parallelization (setting fish$mc.cores = 5). When we set the sets variable to a particular value, this will trigger automatic scoring of hypothesis sets (in this case gene sets). The results are shown in $setres variable.

setname	p	effect	estimate	p.left
TP53 Regulates Transcription of Genes Involved in G1 Cell Cycle Arrest	1.11e-29	16.5 [10.1-27]	16.5	1
RUNX3 regulates CDKN1A transcription	1.24e-29	30.6 [16.9-55.7]	30.6	1
Formation of Senescence-Associated Heterochromatin Foci (SAHF)	2.78e-29	14.2 [8.89-22.5]	14.2	1
Activation of PUMA and translocation to mitochondria	2.93e-29	26.2 [14.8-46.4]	26.2	1
TP53 Regulates Transcription of Caspase Activators and Caspases	3.67e-29	16.9 [10.3-27.8]	16.9	1

Examining the results table, we can see that most of the significant pathways appear related to TP53. Indeed, if we inspect the hypotheses (i.e. genes) contributing to these these gene sets, we will see that they are dominated by TP53 and 1-2 additional genes. For example:

fish$res[sets[[fish$setres[order(p), ][fdr<0.1,  ][3, setname]]], ][order(p), ]

seqnames	start	end	width	gene_name	p	fdr	count	effectsize	count.pred	count.density	count.pred.density	p.neg	fdr.neg
17	7565097	7590856	25760	TP53	6.9e-38	0.00	33	7.72	0.1570	0.029300	0.000139	1.00	1
13	48877887	49056122	178236	RB1	0.00016	0.24	6	3.58	0.5030	0.002180	0.000183	1.00	1
5	126112315	126172712	60398	LMNB1	0.061	1.00	1	1.59	0.3320	0.000768	0.000255	0.95	1
12	66217911	66360075	142165	HMGA2	0.081	1.00	0	-Inf	0.0275	0.000000	0.000137	0.97	1
6	26017260	26018040	781	HIST1H1A	0.24	1.00	0	-Inf	0.2390	0.000000	0.000370	0.80	1
6	27834570	27835359	790	HIST1H1B	0.29	1.00	0	-Inf	0.2020	0.000000	0.000311	0.82	1
6	36644305	36655116	10812	CDKN1A	0.46	1.00	0	-Inf	0.0543	0.000000	0.000140	0.95	1
16	4896666	4932361	35696	UBN1	0.48	1.00	0	-Inf	0.4690	0.000000	0.000138	0.65	1
22	24407642	24574596	166955	CABIN1	0.57	1.00	0	-Inf	0.7160	0.000000	0.000140	0.54	1
6	34204650	34214008	9359	HMGA1	0.59	1.00	0	-Inf	0.0186	0.000000	0.000132	0.98	1
6	26234440	26235216	777	HIST1H1D	0.67	1.00	0	-Inf	0.1890	0.000000	0.000285	0.83	1
22	19318221	19435224	117004	HIRA	0.69	1.00	0	-Inf	0.5410	0.000000	0.000163	0.62	1
22	38201114	38203442	2329	H1F0	0.71	1.00	0	-Inf	0.0719	0.000000	0.000132	0.93	1
6	26156559	26157343	785	HIST1H1E	0.76	1.00	0	-Inf	0.1260	0.000000	0.000355	0.88	1
6	26055968	26056699	732	HIST1H1C	0.88	1.00	0	-Inf	0.2010	0.000000	0.000314	0.83	1
6	119215384	119230332	14949	ASF1A	0.97	1.00	0	-Inf	0.1170	0.000000	0.000275	0.89	1

This is a common challenge with pathway analysis of mutations, since many cancer pathways are usually driven by a single “celebrity” gene.

fishHook Tutorial

Zoran Gajic and Marcin Imielinski